Biopsychosocial predictors and trajectories of work participation after transdiagnostic occupational rehabilitation of participants with mental and somatic disorders: a cohort study

The aim is to study (re)entry to work after occupational rehabilitation in a sick-listed population that is heterogeneous in terms of diagnoses, length of work disability and employment status. Work participation is investigated as it changes over the course of 56 weeks following completion of a common, on-site occupational rehabilitation program. We compare work trajectories for subgroups of participants that were grouped by biopsychosocial factors. The aim is approached through the following research questions:

This cohort study is nested in a randomized controlled study investigating effects of boosted RTW follow-up by telephone. ClinicalTrials.gov. No. NCT01568970 [35]. Reporting is according to STROBE guidelines [39].

Sick listed individuals on temporary work-disability benefits participated in an Acceptance and Commitment Therapy-based occupational rehabilitation program of 3½ weeks duration. Upon admission, they were consecutively invited to participate in the study (January 2012–June 2013). Participants had been referred to the program by general practitioners or other medical specialists and had prior to admission been assessed by an interdisciplinary team consisting of a medical doctor, a psychologist and a physiotherapist.

Participants should be ages 18–59 years old and have a musculoskeletal or other chronic pain disorders, chronic fatigue or a common mental disorder. Further diagnostic criteria were not advised upon referral to allow for inclusion of a real-world population.

Participants should have received sickness certification and be recipients of a temporary medical benefit due to work incapacity (duration over 8 weeks, partial or full-time). By the Norwegian welfare system this involves either sickness benefit (compensation for loss of income) or work assessment allowance (for those who have either already received sickness benefits for the maximum period of 52 weeks, or have not earned the right to sickness benefits through previous work-related income).

Participants should have a self-defined goal of increasing participation in competitive work, be adequately treated for health problems demanding acute care. To participate in the group-based on-site program participants needed to be able to communicate in Norwegian and maintain basic daily care for themselves during a stay at the rehabilitation center.

Exclusion criteria were severe mental illness (ongoing mania, psychosis or suicidal ideation), active substance abuse and addiction, pregnancy, planning to enter/return to studies rather than competitive work, incomplete study registration procedure or not completing the rehabilitation program due to acute injury/disease or family reasons.

Internet-based self-report questionnaires were filled out by participants prior to entering the rehabilitation program. Pre-admission assessment was at an outpatient department of the university hospital while participation in the onsite 3-½ week program took place at an affiliated rehabilitation center. The rehabilitation program was group-based but included individual sessions. Acceptance and Commitment Therapy was an integrated part of mental training, physical training and work-related problem solving. Collaboration with community stakeholders (the general practitioner, the work place, the social security office and others) was initiated on-site and participants prepared a personal action plan for increasing work participation. The on-site program is further described elsewhere [40].

Baseline self-reported data, clinician reported data and registry-based social security data were collected prior to entering the program. The Norwegian Labour and Welfare Service provided data on employment state, working hours and benefits; both medical (sickness benefits, work assessment allowance and permanent disability benefits) and non-medical (unemployment benefits, parenting benefits and social benefits). The procedure for data collection is described in detail elsewhere [35].

Main outcome was (re)entry to the ordinary work force analyzed from baseline and up to 56 weeks after completing the rehabilitation program. The outcome variable was dichotomous and defined as participation in competitive work ≥ 1 day (7.5 h) per week on average over 8 weeks. This is considered to be substantial enough to represent a meaningful “first step” towards entering the ordinary work force and a relevant cut off for individuals with an anticipated poor prognosis for work. For sensitivity analyses we used the outcome measure “minimum half time work” defined as ≥2.5 days (18.75 h) per week on average over 8 weeks. This outcome explores work participation among participants with a stronger connection to the work force. Levels of reentry to work, defined as differences in the estimated probabilities of work participation at all time points after baseline and trend of (re)entry to work, defined as the odds of average reentry over time, from baseline to end of follow-up were analyzed for all grouping factors.

Variables used as grouping factors were selected a priori among expected predictors of RTW [7, 36] and variables of particular interest to the project, for example therapy-related factors thought to be modifiable ‘. psychological flexibility [41]), benefit-related factors (e.g. grading of sick leave) expected to influence work participation [42] and health-related factors (e.g. diagnosis) used in clinical decision support and to inform policy planning in mental health care and work disability management. Predictors of (re)entry to work spanned major biopsychosocial categories and included individual (micro level) and system (meso and macro level) factors [6]. Variables were divided into five main categories:

Variables were dichotomized into clinically recognizable sub-groups to support easy use and meaningful interpretation in the health and social service setting.

Gender, age, level of education and personal economy were self-reported. Age was dichotomized at ≥36 years old since this is a legislative cut-off used in the Norwegian social security system when applying for special disability benefits for young people [43]. The participant’s financial situation was classified using the single item “How do you assess your economy?” The response alternatives are “good, medium, poor”. Participants answering “poor” were registered as having a poor economy.

Registry data on employment state (employed/unemployed), type of benefit (sickness benefit/ work assessment allowance) and grading of sick leave (out of work and on a full benefit/ part-time work combined with a partial benefit) was used. Type of work was classified as physically demanding job (no/yes) based on self-report from The Psychosocial Assessment Instrument (PAI) using one single item: “Is your work very physically demanding?”. The response alternatives are “not at all, a little, quite a bit, much”. A cut-off was set at ≥ “quite a bit”. PAI was developed by two experienced occupational psychologists to address psychosocial factors that could predict the course of somatic symptom disorders.

Psychological flexibility was assessed using the 7-item (short version) of the Acceptance and Action Questionnaire (AAQ-II) [44]. Each item has seven responses ranging from 1(never true) to 7(always true). Higher scores indicate psychological inflexibility and experiential avoidance. Items on the AAQ-II correspond to the processes detailed by the ACT model (action, self-as-context, presence in the moment, values, cognitive defusion and acceptance). The AAQ-II has shown satisfactory structure, reliability and validity [44, 45]. The cut-off was set at > 28, the upper limit of the range considered to indicate a clinically relevant level of distress [44]. Psychological flexibility was of particular interest as a predictor of work since it is amenable to change [36] and specifically targeted through acceptance and commitment therapy [46], the method used in the rehabilitation program.

Expectation of RTW was measured through a single item of the Fear Avoidance Beliefs Questionnaire (FABQ) – Work Subscale [47]. The item used asked “I do not think that I will be back in my ordinary work within three months” with 7 item responses range 0 to 6, with a higher score indicating a worse fear of not returning to work [47]. The cut-off was set at ≥3 indicating a non-positive expectation defined as an “uncertain or even poorer expectation of RTW”.

The Short-Form 8 measures quality of life through eight subscales with one item per subscale [50]. Two of the single item subscales assess role limitations: The Role-Physical (RP) item assesses role limitations due to physical health problems with responses ranging from “Cannot perform daily work at or away from home as a result of physical health” to “No difficulty performing daily work at or away from home as a result of physical health”. The Role-Emotional (RE) item assesses role limitations owing to personal or emotional problems with response ranging from “Cannot perform work, school, or other daily activities due to personal or emotional problems” to “No limitations in performing work, school, or other daily activities due to personal or emotional problems”. Both items capture the impact of role limitations on performing work or other usual activities. Each item has 5 response categories from 1 to 5. The cut-off was set at ≥3 indicating role limitations ranging from a bit to total limitation.

Descriptive data are provided to describe the population at baseline. Regression analysis for repeated measures is used to study trajectories for work participation. Specifically, for each biopsychosocial variable we used generalized estimated equations (GEE) regression analysis to analyze the dichotomous outcome variable (≥ 1 day of competitive work per week) using repeated measurements, an unstructured working correlation structure and treating time as a categorical variable. Each 8-week follow-up period was added to the model as a dummy variable (i.e. weeks 1–8, weeks 9–16, weeks 17–24, weeks 25–32, weeks 33–40, weeks 41–48, weeks 49–56) and the first 8-week period prior to and including time of entry to the program was used as reference category. Precision was measured with 95% confidence intervals (CI). To assess differences between subgroups during follow-up, interaction terms between the studied variable and each registration time-point were included in the model. Estimated marginal values are presented as graphs showing trajectories of work participation.

Distinct levels of (re)entry to work over time within subgroups were assessed for substantial difference in probability for work participation within a subgroup (e.g. employed compared to unemployed) for all time periods after rehabilitation. Substantial differences were indicated by the combination of consistent differences in the estimated probabilities of (re)entry after baseline together with adequate precision of the estimated probabilities at the level of 95% confidence.

All 278 participants admitted to the on-site rehabilitation program were assessed for eligibility. Of these 29 patients were not eligible, 36 patients declined to participate and one participant was lost to follow-up. This left 212 participants to enter the analysis. The flow of participants is reported elsewhere [35].

Baseline characteristics are reported in detail in Table 1.

Trajectories for (re)entering the work force during the first year after the rehabilitation program are presented in (Figs. 1, 2, 3, 4 and 5) with separate graphs for each grouping factor. Each figure represents one of the five categories of biopsychosocial variables.

Participants showed increasing (re)entry to work over time for all of the explored factors, both across and within grouping factors (Fig. 1–5). Lower levels of reentry to work throughout the year correlated with expected negative predictors (e.g. low expectation of return to work) and higher levels of (re)entry correlated with expected positive predictors (e.g. employment and partial sick leave). This same pattern was seen for low precision estimates, except that for health factors consistent level differences were not discernable, perhaps with the exception of mental disorders diagnosed at the level of specialist care. These findings are in accordance with the main body of literature on models and determinants of RTW [7, 60] indicating that expectations of return to work and other work- and system-related factors have greater influence on the process of (re)entry to work than health-related factors. Notably, health factors relating to symptom and diagnosis did not distinguish between those who return to work and those who do not (similar levels and trends of reentry).

For the majority of factors the trend for (re)entry was not significantly different between subgroups. Even for subgroups that showed substantially lower levels of (re)entry, e.g. the unemployed or those with low expectancy of RTW, there is a similar or stronger positive trend (upward pointing slope) of (re)entry throughout the one-year follow-up period. Thus, relative to their poorer starting point these groups in fact showed satisfactory improvement with regard to (re)entry. This is an important distinction and addition to the RTW literature. In clinical practice, negative prognostic factors have commonly been used to disqualify participants from entry to RTW programs, claiming that participants with an assumed poorer prognosis are less likely to draw benefit from programs. Our findings are to the contrary, suggesting that the subgroups of individuals with a weaker starting point (probability of work participation) likewise progressed towards work following transdiagnostic occupational rehabilitation when compared to their counterpart subgroups. The exception was persons on work assessment allowance who did not progress towards work as quickly as their counterparts on sickness benefit.

Work assessment allowance is by legislation a benefit granted to sick listed persons with a weaker connection to the work force, either in terms of lacking or limited previous employment or by having exhausted their right to sickness benefits through sick leave of over 1 years duration. We found a poorer prognosis for (re)entry to work among participants receiving work assessment allowance, seen as both lower level (not substantial) and significantly weaker trend of (re)entry. Sensitivity analysis confirmed that reentry above the “halftime” work (at least 2,5 days per week on average) threshold was substantially lower for those on work assessment allowance than for those on sickness benefit. National reports show high permanent disability and retarded inclusion into the work force for persons on this benefit [61, 62]. Our findings confirm that persons on work assessment allowance, i.e. those with a poor connection to the work force and/or benefits of longest duration, appear to struggle most with (re)entry to work.

Partial sick leave (combination of benefits and paid work) at baseline indicated higher levels of work participation overall, but a poorer trend for increasing entry to work over time, when compared to those on full sick leave. Partial sick leave at baseline seems to protect a relatively stable proportion of the population against falling out of work (at group level), but is associated with an overall slower trend of increasing work participation within the first year. Although not directly comparable, this resonates with previous findings indicating that graded participation supports higher and earlier work inclusion, but may prolong total time on benefits for the group as a whole [63].

Similar trajectories for work participation were seen among participants with mental and non-mental disorders, regardless of source of diagnostic data (self-reported, by sickness certification or by interdisciplinary assessment). I t is well documented that in the general population persons with mental disorders face higher barriers related to work participation and reentry [9, 64, 65]. Surprisingly, the results suggest that for long-term sick leave and after participating in transdiagnostic rehabilitation, pathways for reentry to work are comparable across common mental and somatic disorders. Results should be interpreted with attention to the fact that we have not studied the average population on sick leave, but rather a sample of persons on long-term sick leave that were referred to rehabilitation in specialist care. Work disability was both long lasting and had not been resolved through primary care and community level rehabilitation efforts. In a recent Norwegian government report it was shown that the hazard rate for return to work fluctuated over time in the same way for sick-listed persons with mental and musculoskeletal disorders [66]. This indicated that other employment related factors factors were more important than health factors. Distinguishing between somatic and mental health problems may be less relevant when long-term work disability is the problem.

With regard to functioning, participants that experienced role limitations due to emotional problems at baseline had stronger trend for reentry than those not experiencing such. Hypothetically, this could be due to improved emotional functioning following the transdiagnostic intervention, as a result of the intervention and/or the return to work process itself. Future studies may wish to focus on this and other potentially modifiable factors that can be targeted in therapy (e.g. emotional functioning, return to work expectations and psychological flexibility).

The findings from this study fit with an emerging body of evidence [60] suggesting that persons with long term work disability due to common health problems such as common mental disorders and musculoskeletal/pain disorders may best be managed with a holistic and less diagnose-focused approach to occupational rehabilitation. Similarities in etiology, assessment and therapeutic approach that cross diagnostic boundaries have been pointed out [6, 7, 29, 67]. Most studies persist in recruiting participants from relatively homogenous samples and dichotomizing between psyche and soma. Traditional clinical boundaries and the researchers preference of standardized setting to support internal validity are adhered to. However, the problem of generalizability arises, with lack of external validity and subsequent problems with implementation in ordinary practice [68]. Various approaches are recommended when developing complex interventions in real world contexts [69]. This paper adds to the literature on occupational rehabilitation by giving a different perspective on prediction analysis than the preferred multivariable prediction model.

Prediction of outcome for occupational rehabilitation has historically primarily been based on clinical judgment of factors that are considered relevant. The range of factors has depended on the applied conceptual model (e.g. biomedical, biopsychosocial, etc.). To increase precision empirical evidence has been applied to create actuarial models. The gold standard is the multivariable model, but some weaknesses have been pointed out [70, 71]: Variables are assumed stable and static. Misclassification is risked if there is no room for individual difference. Problems arise if critically important data are not collected or if important variables are imprecisely measured. Many prospective studies are of limited scope and few studies actually cover the requisite range of variables for comprehensive prediction analysis. Over fitting of the model may occur and result in reduced external validity. Generalizability to other populations and contexts is often unknown. Conclusively, actuarial models may not be easily translated into secondary prevention applications. In clinical settings and for policy planning in broad populations, simpler and less conservative prediction models may be more useful and recognizable.

We anticipated that a strict actuarial model might not give reliable results while other possible important associations could be lost [70]. In such situations methodological compromise is recommended. In this study, exploratory analysis using multiple adjusted single factor models was the method of choice. This non-multivariate model is less conservative and has limitations, but was chosen to match the research questions, support generalizability [72] and acknowledge the challenges of complex transdiagnostic interventions in real world research. This simpler model translates for easy use in clinical practice and supports comparison of closely correlated variables of practical interest (e.g. mental disorders diagnosed in different settings). Furthermore, results from single factor prediction models can be used to generate more complex, computer-based prediction models using big data.

This study adds new understanding of the work (re)entry patterns of a complex, real-world population. This knowledge may generate further hypothesis, and be used in developing decision support for general practice and other health and welfare services involved in planning occupational rehabilitation for heterogeneous populations. Prediction rules are typically developed in rather homogenous samples but are practically useful only if they provide accurate risk prediction in different settings [77]. Guidelines for work disability management are recommended further developed with attention to conceptual models that are applicable across health conditions (22, 36).

Our results show that groups that are often viewed as marginalized from the work force (e.g. those with a mental disorder or the unemployed) may progress towards work as much or more after the same transdiagnostic intervention as groups that do not face the same barriers. These marginalized groups showed the same or higher potential for reentry to work (similar or stronger trend) supporting that they should at minimum be offered equal opportunities within work disability management.

Persons at high risk of not being included in the work force may benefit from a transdiagnostic approach that draws attention to work place and social security issues. This is in line with recommendations from other studies within the field of occupational rehabilitation. A related qualitative study on the studied intervention recommended a strengthened focus on specific steps towards return to work [78]. Other RTW interventions have shown that adding workplace strategies may provide enriched work environments that support the underlying therapeutic process [79‐81]. This study does not answer how work and social security issues can best be dealt with and future research is needed. Cautionary sub-group analysis from the randomized controlled trial carried out alongside this study indicated that groups with lower probabilities of entering work may benefit from a simple regime of telephone support in the phase of reentering work and daily life [35]. An ongoing randomized controlled trial will study the effect of adding a workplace component to the intervention [82].

This transdiagnostic group-based occupational rehabilitation intervention has shown promising feasibility for implementation and acceptability to participants [26] but needs to be developed further. This is of interest since common rehabilitation programs for mixed groups of participants potentially have financial, practical and therapeutic benefits. Inference cannot be drawn from this study on the effect of the transdiagnostic occupational rehabilitation program, and further research is needed to establish for which persons a transdiagnostic approach to occupational rehabilitation may be preferred, and how transdiagnostic programs should be personalized.